power control for underlay cognitive radio networks with
TRANSCRIPT
Power Control for Underlay Cognitive Radio Networks with Full-duplex
Transmissions
by
Ningkai Tang
A thesis submitted to the Graduate Faculty ofAuburn University
in partial fulfillment of therequirements for the Degree of
Master of Science
Auburn, AlabamaMay 4, 2014
Keywords: Full-duplex Transmissions, Cognitive Radio, PID Controller
Copyright 2014 by Ningkai Tang
Approved by
Shiwen Mao, Chair, McWane Associate Professor of Electrical and Computer EngineeringPrathima Agrawal, Ginn Distinguished Professor of Electrical and Computer Engineering
John Hung, Professor of Electrical and Computer EngineeringJitendra Tugnait, James B. Davis and Alumni Professor of Electrical and Computer
Engineering
Abstract
Radio communication, one of core contents in wireless communication, is facing a great
crisis of running out of resource. Unlike other energy crisis which can be solved by using all
kinds of alternative energy, the only resource of radio communication is channel, in another
word, spectrum. Rather than sticking on trying to achieve the rest one percent of capacity,
more and more scholars start to realize that methods reuse the existing available channel
such as cognitive radio and full-duplex transmission lead a much brighter future.
Both cognitive radio (CR) and full duplex transmissions are effective means to enhance
spectrum efficiency and network capacity. In this paper, we investigate the problem of
power control in an underlay CR network where the CR nodes are capable of full-duplex
(FD) transmissions. The objective is to guarantee the required quality of service (QoS) in
the form of a minimum signal-to-interference-plus-noise (SINR) ratio at each CR user and
keep the interference to primary users below a prescribed threshold. We design an effective
distributed power control scheme that integrates a proportional-integral-derivative (PID)
controller and a power constraint mechanism to achieve the above goals. We analyze the
stability performance of the proposed scheme and develop a hybrid scheme that can switch
between FD and half duplex modes. The proposed schemes are validated with extensive
simulations.
ii
Acknowledgments
In the beginning, I would like to express my deepest thanks to my committee chair and
advisor Dr. Shiwen Mao, who has guided me over wireless communication area throughout
my master life and inspired me of this idea. Without his support, I could have no chance to
learn so much knowledge, not to mention to behave like a real researcher.
I also would like to thank all the other committee members of mine. Dr. Prathima
Agrawal has introduced a lot of new knowledge to me in her seminar; Dr. John Hung has
shown me the basic concept of control theory which is very crucial to my thesis; And Dr.
Jitendra Tugnait has get me through with lots of stochastic and channel problems. Sincerely
thank you, for I could never have accomplished pursuing my master degree here.
Then, I need to thank all the other professors who has taught me in Auburn University:
Dr. Fa Dai, Dr. Bogdan Wilamowski and Dr. Chwan-Hwa Wu. Your knowledge has indeed
broadened my view. Meanwhile, I need to thank Dr. Soo-Young Lee, without his help I
would probably lose the opportunity to defense in time.
In addition, I want to take this opportunity to appreciate the friendship and support
from all my fellow colleagues in the Electrical and Computer Engineering at Auburn Uni-
versity: Jing Ning, Yi Xu, Dr. Hui Zhou, Yu Wang, Zhifeng He, Zhefeng Jiang, Yu Wang
(Same name), Xuyu Wang and Mingjie Feng. In addition, I want to thank our senior alumni
Dr. Yingsong Huang for the knowledge he has generously offered me during my research.
Finally, I would like to thank my dear parents and wife for their understand and support
all these years, I really owe you a lot. I would also like to thank my future daughter, who is
my strongest motivation during these days. I love them all forever!
iii
This work is supported in part by the US National Science Foundation (NSF) under
Grants CNS-0953513, and through the NSF I/UCRC Broadband Wireless Access & Ap-
plications Center (BWAC) site at Auburn University (Grant IIP-1266036). Any opinions,
findings, and conclusions or recommendations expressed in this material are those of the
author(s) and do not necessarily reflect the views of the NSF.
iv
Table of Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Cognitive Radio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Software-defined Radio . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 The Birth and Definition of Cognitive Radio . . . . . . . . . . . . . . 2
1.1.3 Interoperability and Dynamic Spectrum Access . . . . . . . . . . . . 3
1.2 Full-duplex Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.1 Definition of Full-duplex Wireless Transmission . . . . . . . . . . . . 9
1.2.2 Problem and Ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Motivation for Our Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Background and Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 System Model and Problem Formulation . . . . . . . . . . . . . . . . . . . . . . 15
2.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Terms and Parameter Definition . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Signal-to-Interference-plus-Noise Ratio . . . . . . . . . . . . . . . . . 17
2.2.2 Additive White Gaussian Noise Channel . . . . . . . . . . . . . . . . 17
2.2.3 Channel Capacity and Transmission Rate . . . . . . . . . . . . . . . . 17
2.2.4 Fading Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.5 Self-Interference Suppression Factor . . . . . . . . . . . . . . . . . . . 18
2.2.6 Mode Tuning Threshold . . . . . . . . . . . . . . . . . . . . . . . . . 18
v
2.3 Problem Statement and Equations . . . . . . . . . . . . . . . . . . . . . . . 18
3 Power Adjustment Schemes and System Analysis . . . . . . . . . . . . . . . . . 23
3.1 PID Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Power Adjustment Constraint . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Analysis of Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 HD-FD Tuning Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4 Simulation and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1 Simulation Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Control Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.3 Throughput Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5 Conclusions and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
vi
List of Figures
1.1 SDR product [2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Safety teams with different standard or operating frequency [22] . . . . . . . . . 4
1.3 Platform for inter-radio system switching with CR [10] . . . . . . . . . . . . . . 5
1.4 Frequency allocation charts for the United States [18] . . . . . . . . . . . . . . . 6
1.5 Spectrum measurement across the 928 to 948 MHz band on June 19, 2008
(Worcester, MA, USA) [22] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6 Full-duplex wireless transmission . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Half-duplex wireless transmission . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.8 SIS design proposed in Mayank Jain et. al . . . . . . . . . . . . . . . . . . . . . 12
2.1 An FD underlay CR network considered in this paper. . . . . . . . . . . . . . . 16
3.1 The PID controller design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 System control block diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1 Distribution of the TRs and DPs in the 100×100 network. . . . . . . . . . . . . 32
4.2 Evolution of the SINR at TR5 when Γ = 0.1 and χ = 0.00005. . . . . . . . . . . 33
4.3 Evolution of the transmit power at TR5 when Γ = 0.1 and χ = 0.00005. . . . . . 33
vii
4.4 Power adjustment u5(t) = min{z5(t), c5(t)} of TR5 when Γ = 0.1 and χ = 0.00005. 34
4.5 PID controller adjustments z5(t) when Γ = 0.1 and χ = 0.00005. . . . . . . . . . 35
4.6 PU protection constraint adjustments c5(t) when Γ = 0.1 and χ = 0.00005. . . . 35
4.7 Maximum measured interference among all the DPs when Γ = 0.1 and χ = 0.00005. 36
4.8 Maximum measured interference among all the DPs when Γ = 0.3 and χ = 0.00005. 37
4.9 Evolution of the SINR at TR5 when Γ = 0.3 and χ = 0.00005. . . . . . . . . . . 37
4.10 Evolution of the SINR at TR5 under varying Γ. . . . . . . . . . . . . . . . . . . 39
4.11 Evolution of the SINR at TR5 under varying Γ. . . . . . . . . . . . . . . . . . . 39
4.12 Maximum measured interference among all the DPs under varying Γ. . . . . . . 40
4.13 Distribution of the TRs and DPs in the 1000×1000 network. . . . . . . . . . . . 40
4.14 Average throughput of the FD, HD, and hybrid modes for different values of χ. 41
4.15 Average throughput of the FD, HD, and hybrid modes for different values of σ2. 42
viii
List of Tables
4.1 Setting of Γ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
ix
Chapter 1
Introduction
1.1 Cognitive Radio
In recent years, an unprecedented increase in wireless data has been observed, largely
due to the proliferation of smartphones, tablets and other wireless devices. The exploding
wireless data calls for effective technologies for enhancing spectrum utilization and wireless
network capacity. To this end, cognitive radios (CR) have been recognized as one of the
key technologies to meet this grand challenge on wireless network capacity. As an effective
means of sharing spectrum among licensed (i.e., primary) users (PU) and unlicensed (i.e.,
secondary) users (SU), CR has been demonstrated to achieve high utilization of the scarce
spectrum resource [23, 24].
Due to the rapidly running out of spectrum over available frequency in communication,
CR has become a promising solution to reuse frequency resource. CR was motivated by
recent spectrum measurements by the Federal Communication Commission (FCC), where
temporal and geographical variations in the utilization of assigned spectrum are found to
range from 15% to 85%, and a significant amount of the spectrum remains unutilized. A
CR is an advanced radio device that enables dynamic spectrum access (DSA). It represents
a paradigm change in spectrum regulation and access, from exclusive use by primary users
to shared spectrum and dynamic access for secondary users, to enhance spectrum utilization
and achieve high throughput capacity. CR has profound impact on how future wireless
networks will be designed and operated and has become one of the suggested key concept of
5G criterion.
1
Figure 1.1: SDR product [2]
1.1.1 Software-defined Radio
Before saying something about CR, we need to acquire the concept about one of its key
predecessor technology software-defined radio (SDR) first.
With the development of wireless communication and microelectronics technology, SDR
has been officially introduced to the public in 1991. With its presence, it is defined as “A
radio platform of which the functionality is at least partially controlled or implemented in
software” [22]. According to the definition, if certain kind of waveform has been saved in
the memory of SDR product, it should be able to be employed on any frequency.
Thanks to the industriousness of manufactures, the hardware needed for SDR has be-
coming quite affordable for commercial use. This also induced more attention about this
technology and caused a virtuous cycle. Nowadays, SDR product like what is shown in
Fig. 1.1 is quite common to see and benefits a lot of ensuing works.
1.1.2 The Birth and Definition of Cognitive Radio
Joseph Mitola III [1], who is now a professor at Stevens Institute of Technology, is
well known as the father of CR technology. In 2000 he finished his doctoral defense with
2
dissertation “Cognitive radio—An integrated agent architecture for software defined radio”
and defined CR for the first time.
At first, CR is seen as a “intersection of personal wireless technology and computational
intelligence” and should be defined as “A really smart radio that would be self-aware, RF-
aware, user-aware, and that would include language technology and machine vision along
with a lot of high-fidelity knowledge of the radio environment”. However, just like Hamlet
in eys of different people, this definition is too large for a single technology and has caused
various understandings. In “Cognitive radio: Brain-empowered wireless communication”
[11], Simon Haykin redefined CR as “an intelligent wireless communication system that
is aware of its surrounding environment, and uses the methodology of understanding-by-
building to learn from the environment and adapt its internal states to statistical variations
in the incoming RF stimuli by making corresponding changes in certain operating parameters
in real-time, with two primary objectives in mind: highly reliable communications whenever
and wherever needed; efficient utilization of the radio spectrum.” To simplify the conclusion,
he used six words to represent the characteristcs of CR: “awareness, intelligence, learning,
adaptivity, reliability, and efficiency”.
Due to the blossom of CR research, U.S. FCC has proposed their official and strict
definition of CR which is “A Cognitive Radio is a radio that can change its transmitter
parameters based on interaction with the environment in which it operates. The majority
of cognitive radios will probably be SDR but neither having software nor being field pro-
grammable are requirements of a cognitive radio”. From then on, this definition becomes a
recognized meaning for CR.
1.1.3 Interoperability and Dynamic Spectrum Access
For its all kinds of capabilities, CR has enabled a list of applications. However, the most
well-known and widely used two implementations are Interoperability and DSA.
3
Figure 1.2: Safety teams with different standard or operating frequency [22]
Interoperability
How to understand interoperability? Let us start from a existing tragedy first: In August
2005, The Hurricane Katrina struck New Orlean, caused mass casualty and property loss.
After the disaster, a lot of people has blamed government for reacting too slow. Not everyone
knows that, but one important reason for the problem is the lacking of interoperability. The
communication equipment used by local rescue teams (For instance, police and hospital) has
different specifications so that they cannot cooperate with each other efficiently, like Fig. 1.2.
Imagine, if they can communicate with each team smoothly, how many lives can be
saved? Although set their equipments all in a united frequency at the beginning is not
possible due to feasibility and security problems, we could still solve the problem by benefiting
from interoperability of CR. As we have discussed in the former parts, CR has the ability of
adaptivity and is fully capable of dealing this. CR may reconfigure all the existing standards
in a single form, thus to allow them communicating with everyone.
4
Figure 1.3: Platform for inter-radio system switching with CR [10]
Interoperability even have other practical use in commercial usage. Fig. 1.3 shows the
design of inter-radio system with cognitive radio system. In the figure, a cognitive radio
basestation has the ability to “translate” all kinds of wireless standards so users will be able
to connect with each other even they used to be in a different network. Still fearing about
phones cannot be used in a foreign country? That is an old story.
Dynamic Spectrum Access
Spectrum is now considered as a valuable resource in wireless communication, and re-
source is always accompanied with allocation problem. International Telecommunication
Union (ITU) has provided a general plan for global spectral use and FCC is in charge of
regulating spectrum use in the U.S.. As shown in Fig. 1.4, spectrum is allocated mostly for
licensed users. Licensed user pays considerable money for occupying certain band in order
to guarantee their QoS.
5
Figure 1.4: Frequency allocation charts for the United States [18]
6
Figure 1.5: Spectrum measurement across the 928 to 948 MHz band on June 19, 2008(Worcester, MA, USA) [22]
As we all know, spectrum is limited, with most of the spectrum occupied by licensed
user, the usage of radio communication seems to be “dead” too. However, the efficiency
of the current allocation mechanism has been challenged by more and more surveys. As a
matter of fact, these surveys indicate considerable waste in the usage of licensed spectrum,
like Fig. 1.5. Frequency around 800Hz is usually seen as one the best range since it is both
good for wall-piercing and transmission. Knowing the value of such spectrum forces people
rethink about current allocation.
Upholding the idea of saving and basing on the possible technology, FCC prompts
the concept of DSA. DSA encourage unlicensed user to “borrow” spectrum from licensed
users. Quite obvious, the CR platform is facing two problems here: The needing of being
environmentally aware and the needing of rapidly reconfigurable. With the progress of CR
technology, the idea of DSA which needs unlicensed user accessing to licensed spectrum while
7
protecting primary user (PU) is completely realizable. For a secondary user (SU), usually
there are two approaches to accomplish such task: Overlay and underlay.
In overlay approach, CR will try to find out some random, small idle bands rather than a
single wode spectrum. This shows the “aggregate” ability of overlay CR. Figuratively, overlay
CR is working like pouring sand into the crevice of stones. Stones are like SUs who has filled
in the space, and sand like overlay SUs can fully utilize the rest resource. The main idea of
overlay CR technology is to reuse all the unused spectrum, since there is no PU using it, it
should never collide with PU. In another word, it is temporarily playing the role of a PU. One
representative application for overlay is opportunistic spectrum access (OSA), which usually
goes hand in hand with multicarrier modulations schemes and spectrum sensing technology.
Overlay may sometimes offense the right of PUs due to false spectrum sensing. Although
overlay is solving the problem, it sometimes has to face the pain of lacking stability.
To the opposite, underlay uses another way of thinking. While overlay is surely a great
idea and is making good use of wasted spectrum, underlay is even astonishing and shows
the ability of utilizing spectrum which is still in use. Rather than the sand in the crevice,
underlay technology is more like the cream on a cup of Mocha. Coffee and cream stays
in different layer, they never collide with each other and can stay together in peace. To
achieve such characteristics, underlay technology has to apply power control technology and
remain in a low interference. Underlay technology is capable of using any band due to its
characteristic, however it also has problems like being unable to transmit in long distance
due to low transmission power and its high frequency carrier wave.
In CR networks, the most important design factor is to balance the tension between
PU protection and SU spectrum access gains [23]. On one hand, the capacity of SUs should
be maximized to “squeeze” the most out of the spectrum. On the other hand, the adverse
impact to PUs, resulting from sharing spectrum with SUs, should be kept below a tolerable
level. Obviously, these are two conflicting goals that should be balanced in the design of
CR networks. In the so-called overlay CR networks, PU protection is achieved by spectrum
8
Figure 1.6: Full-duplex wireless transmission
sensing and spectrum access only when the PUs are sensed absent [23]. In the so-called
underlay CR networks, both PU and SU transmissions coexist in the same spectrum band,
and PU protection is achieved by carefully controlling the power of the SU transmitters [17].
1.2 Full-duplex Transmission
Recently, a breakthrough in wireless communications is FD transmission [5–7,14]. Tra-
ditionally, wireless communications are all half duplex (HD) due to the large path loss
typical in wireless transmissions. If FD transmission is allowed, the self-interference will be
so strong (like the sun) and the weak received signal from a remote transmitter (like stars)
will be completely overwhelmed and cannot be decoded. Recently, encouraging results have
been reported on enabling FD wireless transmissions in both single link and a network set-
ting [5–7, 14]. The enabler of HD is the recent advances in self-interference suppression
(SIS). Various effective SIS techniques have been proposed and tested, such as antenna
separation [7], antenna cancellation [6], signal inversion and adaptive cancellation [14], and
combined optimal antenna placement and analog cancellation [20]. In [20], the author showed
a practical implementation that can suppress self-interference (SI) for up to 80 dB, which
should be sufficient for many application environments [3].
1.2.1 Definition of Full-duplex Wireless Transmission
FD wireless transmission is defined as a wireless transmission which is communicating
in both directions simultaneously, as shown in Fig. 1.6.
9
Figure 1.7: Half-duplex wireless transmission
FD is quite common in wired communication equipments, such as Ethernet. The tech-
nology is quite simple too: It simply enables connections work by making simultaneous use
of two physical pairs of twisted cable, one pair for receiving packets the other is used for
sending packets. However, when things come to single channel wireless network, everything
has changed.
1.2.2 Problem and Ideas
Until today, most wireless equipments still remains in the half-duplex (HD) age like
Fig. 1.7 Unlike the wired communication, wireless has some innate problem in implementing
FD on a single channel: The direction of signal cannot be guided. For a wireless equipment,
the self-issued signal is always omnidirectional and will also be received by the receiver of
its own, with little path loss due to extremely near distance. Meanwhile, the desired signal
coming from the other equipment is nearly negligible comparing to the signal transmitted
by itself. As we may all know, a signal with too low SINR could not be decoded. And that
is the problem of FD on a single channel: SI adds a destructive impact on its own receiving
antenna.
10
After locking on the problem, some idea to solve the problem has already emerged, and
they are all classified into SIS technology.
Radio Frequency Cancellation
Radio Frequency (RF) cancellation is a chip based SIS method [19]. Since we must
know transmitted signal and received signals. So we could use them as inputs and take
the difference of the received signal with the SI (the transmitted signal) as the output. By
changing the amplitude and phase of the interference reference signal we can try to match
the interference in the received signal. An RF splitter is also used to give the transmit signal
to the cancellation circuit as the interference reference.
Digital Cancellation
Digital cancellation is another key technology for SIS. As for now, there are two well-
known way for digital cancellation. First one is decode and cancellation. The transmitted
packets are all marked with special symbol first. After receiving the original signal, the
receiver decodes it, pick out all interference packets and then the packets are clean. The
other digital cancellation is called coherent detection, which use a detector to correlates the
incoming signal with the transmitted signal. Since the detector have full knowledge about
the transmitted signal, it is capable to estimate the delay and phase shift of the received
signal, so it can use the transmit signal to correlate with the interference. The second method
need no modulation of the signal and is backwards compatible so this technique is somehow
seen as a better one.
Antenna Cancellation
Antenna cancellation is maybe the most important technique of the three because it re-
ally has shown a significant ability in SIS. The idea of this method is using pairs of transmit-
ters to destroy their effect on the receiver. As we all know, radio is a kind of electromagnetic
11
Figure 1.8: SIS design proposed in Mayank Jain et. al
wave and it has crest and trough, so by properly placed we can add crest with trough on the
receiver’s location so they will cancel each other. Although efficient, antenna cancellation is
usually deployed together with other two method to ensure a better result, like the design
in 1.8.
1.3 Motivation for Our Work
The high potential of FD has attracted substantial interest. However, the mainstream
FD research nowadays has only focused on feasible physical layer techniques or the per-
formance analysis for certain utilization. Although some advances have been made, the
important problem of guaranteeing overall system performance is yet to be studied. Since
the compelling objective of FD transmission for fully utilizing the limited resource of wire-
less networks meets the core purpose of CR technique, we propose to combine FD with CR.
Thus, certain problems such as “If FD transmission could also improve the performance in
CR networks under certain circumstances?” remains to be answered . Unlike HD trans-
mission which mostly cares about distance as the fading parameter, FD transmission has to
take more consideration of a set of much more complex data such as SI suppression factor
12
and SINR ratio. Their relationship and impact are highly important for our research, and
could lead a revolution to the new communication criterion.
1.4 Background and Related Work
FD transmission is a new technology to push the limit of single channel communications.
In [6], the authors proposed basic concepts such as RF and digital cancellations and discusses
potential MAC and network gains with full-duplexing. In [20], the authors presented the
design and implementation of a real-time 64-subcarrier 10 MHz full-duplex OFDM physical
layer, and demonstrated up to 80 dB SI suppression with experiments. In [14], the authors
presented a full duplex radio design using signal inversion and adaptive cancellation, as well
as a full duplex MAC design and evaluation results with a testbed of 5 prototype FD nodes.
In [5], a MIMO FD design was presented, while FD cellular networks have been investigated
in some recent papers [8, 21].
CR has been recognized as an important technology for enhancing spectrum access
efficiency [23, 24]. In the class of overlay CR networks, SUs sense the spectrum and access
the spectrum when PUs are absent. In the class of underlay CR networks, SUs coexist with
PUs in the same spectrum conditioned on limited interference to the PUs. Both techniques
can be transparent to PUs [23]. In a recent work [3], the authors proposed to combine FD
with CRs. The FD capability can be utilized to allow current two-way transmissions for the
SUs, as well as enabling SUs to transmit while sensing.
Feedback control has found wide application in communication and networking systems.
A modern overview of functionalities and tuning methods for PID controllers was presented
in [4]. In [12], a proportional (P) controller was developed for streaming videos to stabilize
the received video quality as well as the bottleneck link queue, for both homogeneous and
heterogeneous video systems. A modern overview of functionalities and tuning methods
for PID controllers was presented in [4]. In [9], the author presented a PID based power
13
adjustment algorithm that was later extended in [17], which developed a PID control for
power control in underlay CR networks.
In this paper, we investigate several control model based on single channel full duplex
cognitive radio, which is different from previous work. By taking interference cancellation
factor into consideration we can surely acquire a better throughput and more accurate conse-
quence. This work is inspired by the recent idea about single channel full duplex transmission
and has utilized the model listed in [17]. Some math proofs may come after the previous
work which has been finished in [16] and [3].
14
Chapter 2
System Model and Problem Formulation
2.1 System Model
Consider an underlay CR network as illustrated in Fig. 2.1. There is a primary network
with active transmissions using a licensed spectrum band. A co-located secondary network
consists of (s + 1) secondary users (SU), termed TRi, i = 1, 2, · · · , s + 1, where s is an
odd number. The SUs are paired to form (s + 1)/2 FD transmission links, i.e., TRi is
transmitting to, and simultaneously receiving from TRi+1, while i is an odd index. Due to the
underlay spectrum sharing policy, the SUs are allowed to use the same spectrum band as the
primary network. For protection of the primary network, there are p detection points (DP)
in the primary work that measure the interference from the secondary transmissions. Such
interference should be kept below a threshold at the DP locations by effectively controlling
the power of the secondary transmitters.
In Fig. 2.1, gij denotes the channel gain from TRi to DPj; hij represents the channel
gain from TRi to TRj ; and σ2i is the sum of the total interference from primary transmissions
and the noise power at TRi. To simplify notation, we assume channel reciprocity, i.e., hij
(or gij) is equal to hji (or gji) for all i, j.
For each FD link, the SI is Pi(t)h2ii, where Pi(t) is the transmit power of TRi and hii
is the channel gain from TRi’s transmitting antenna to the receiving antenna. We assume
that each TRi utilizes SIS, and the residual SI is reduced to χPi(t)h2ii, where χ is a constant
in [0, 1] depending on the specific SIS design. When χ = 0, it is the perfect case where
the SI can be completely canceled; when χ = 1, it is the worst case without SIS and FD
transmission is not possible. Usually χ is a small number, e.g., at least 45 dB across a 40
MHz band and up to 73 dB for a 10 MHz OFDM signal [14].
15
2
s
σ
g�1 g�2
g1� g�� g��
g�� g� ,�+1 g� g� ,+1
Detection point 1
TR1 TR2 TR� TR�+1 TR TR+1
h,+1 = h+1, 2
1+sσ
h h+1,+1
2
i
σh�,�+1 = h�+1,�
2
1+iσh�� h�+1,�+1
2
1
σh12 = h21
2
2
σh11 h22
h�1 h�2 h� � h� ,�+1 h� h� ,+1
Detection point j Detection point p... ...
Figure 2.1: An FD underlay CR network considered in this paper.
16
2.2 Terms and Parameter Definition
Since this theis contains a lot of terms and parameters, this section aims to clarify the
confusion for what each term or parameter stands for.
2.2.1 Signal-to-Interference-plus-Noise Ratio
Signal-to-interference-plus-noise ratio, or SINR is defined as the power of a certain signal
of interest divided by the sum of the interference power and the power of noise. According
to the value of interference power and noise power, SINR may also stand for Signal-to-
interference ratio (SIR) or Signal-to-noise ratio (SNR).
In the thesis, we use γ to represent SINR.
2.2.2 Additive White Gaussian Noise Channel
Additive white Gaussian noise channel, as know as AWGN channel, express a certain
kind of channel model. Here, the additive noise has a uniform power across the frequency
band and is normal distributed in the time domain.
In the thesis, we assume the whole system is using AWGN channel.
2.2.3 Channel Capacity and Transmission Rate
Channel capacity is defined as the upper bound on the rate of information that can
be reliably transmitted over a communications channel. According to Shannon, the channel
capacity of AWGN channel is
C = Blog2(1 + SNR) (2.1)
Here, B is the bandwidth of channel.
17
In the thesis, we use R to represent the transmission rate, the transmission rate is ideal
rate when there is no error possibility during transmission and has been normalized with
B = 1Hz.
2.2.4 Fading Distance
Since wireless signal will lose power during transmission, we usually use maths model
with distance to evaluate the path loss. The distance is what we call fading distance.
In the thesis, we use dij or di,j to represent the fading distance.
2.2.5 Self-Interference Suppression Factor
In FD technology, the transceiver is capable to suppress its own transmission signal in
the receiving antenna in order to enable FD transmission. However, the transmission signal
cannot be cancelled completely so we use a SIS factor to denote the capability of suppression.
It is defined as the power of processed SI divided by the power of orignial SI.
In the thesis, SIS factor is written as chi.
2.2.6 Mode Tuning Threshold
In our design, we propose a hybrid mode which would select HD or FD mode automat-
ically in order to gain optimal throughput. The factor that decide the mode selection is the
mode tuning threshold.
In the thesis, mode tuning threshold is represented with T .
2.3 Problem Statement and Equations
For the FD CR network to work properly, two conditions should be satisfied by control-
ling the transmit power of the TRj ’s. The first condition is primary user protection. That is,
the measured interference from secondary transmissions should be kept below a prescribed
tolerance level Dj at each DPj . The second condition is guaranteeing the QoS of SUs. That
18
is, the SINR at the TRi’s should be kept above a prescribed threshold Γ, such that the SUs
can be guaranteed with a minimum data rate.
We assume that time is slotted. To achieve these goals, in each time slot t, a centralized
power control algorithm updates the transmit power of each TRi, denoted as Pi(t), according
to the measured radio environment, as
Pi(t+ 1) = Pi(t) + ui(t), (2.2)
where ui(t) is the increment (positive or negative) of power at TRi in time slot t.
We assume that the DPs can detect the interference from the SUs. For example, if the
channel gains and the transmit powers from the primary transmitters are known, the DP
can estimate the interference from primary transmissions. Alternatively, a quiet period as
in IEEE 802.22 WRANs could be enforced for the SUs [13]. Since there is no secondary
transmissions in the quiet period, the DPs can measure the interference from primary trans-
missions. Once the primary interference is known, a DP can estimate secondary interference
by subtracting the primary interference from the total interference it receives.
As shown in Fig. 2.1, the total interference from the TRi’s to a detection point DPj is
yj(t) =s+1∑
k=1
Pk(t)g2kj, j = 1, 2, · · · , p. (2.3)
Then the primary user protection constraint becomes
yj(t) ≤ Dj , j = 1, 2, · · · , p. (2.4)
For time slot t + 1, the secondary interference yj(t + 1) caused by the updated transmit
powers should also satisfy (2.4), i.e.,
yj(t+ 1) ≤ Dj, j = 1, 2, · · · , p. (2.5)
19
For the second constraint on guaranteeing the QoS of SUs, the SINR at the receiving
antenna of TRi can be written as
γi(t) =
Pi+1(t)h2i+1,i
∑s+1
j=1,j 6=iPj(t)h2
ji+χiPi(t)h2ii+σ2
i (t), i is odd
Pi−1(t)h2i−1,i
∑s+1
j=1,j 6=iPj(t)h2
ji+χiPi(t)h2ii+σ2
i (t), i is even,
(2.6)
where χj is the SIS factor [3].
Recall that ui(t) = Pi(t + 1) − Pi(t). From the control point of view, (2.6) can be
regarded as the state equation and ui(t) the input. The updated state is
γi(t+ 1) =
γi(t) +h2i+1,i
Ii(t)ui(t)+
Pi(t+1)h2i+1,i[Ii(t)−Ii(t+1)]
Ii(t)Ii(t+1), i is odd
γi(t) +h2i−1,i
Ii(t)ui(t)+
Pi(t+1)h2i−1,i[Ii(t)−Ii(t+1)]
Ii(t)Ii(t+1), i is even,
(2.7)
where
Ii(t) =
s+1∑
j=1,j 6=i
Pj(t)h2ji + χiPi(t)h
2ii + σ2
i (t). (2.8)
It is shown that generally Ii(t)− Ii(t+ 1) is much smaller than Ii(t)Ii(t+ 1) [15]. It follows
that (2.7) can be approximated as
γi(t + 1) =
γi(t) +h2i,i+1
Ii(t)ui(t), i is odd
γi(t) +h2i,i−1
Ii(t)ui(t), i is even.
(2.9)
Let Γ denote the minimum required SINR for SU TRi. The SU QoS constraint is
γi(t) ≥ Γ. (2.10)
20
The updated γi(t + 1) should also satisfy condition (2.10), i.e.,
γi(t+ 1) ≥ Γ. (2.11)
Define parameters a and b as
a =
[ui(t) + Pi(t)]h2i+1,i, i is odd
[ui(t) + Pi(t)]h2i−1,i, i is even.
(2.12)
b =
s+1∑
j=1,j 6=i
[uj(t) + Pj(t)]h2ji + χi[ui(t) + Pi(t)]h
2ii+
σ2i (t+ 1), i = 1, 2, · · · , s+ 1. (2.13)
From (2.2), (2.5), and (2.11), we derive the following system of equations that can be solved
for ui(t).
a/b = Γ, i = 1, 2, · · · , s+ 1
[ui(t) + Pi(t)]g2ij +
∑s+1k=1,k 6=i[uk(t)+
Pk(t)]g2kj ≤ Dj , j = 1, 2, · · · , p.
(2.14)
If the channel gains vary over time (e.g., in a mobile SU network), we can defined parameters
a∗ and b∗ as
a∗ =
[ui(t) + Pi(t)]h2i+1,i(t+ 1), i is odd,
[ui(t) + Pi(t)]h2i−1,i(t+ 1), i is even.
(2.15)
b∗ =s+1∑
j=1,j 6=i
[uj(t) + Pj(t)]hji(t+ 1)2 + χi[ui(t)+
Pi(t)]hii(t + 1)2 + σ2i (t + 1), i = 1, · · · , s+ 1. (2.16)
21
A similar system of equations can be solved to determine ui(t) as
a∗/b∗ = Γ, i = 1, 2, · · · , s+ 1
[ui(t) + Pi(t)]g2ij(t + 1) +
∑s+1k=1,k 6=i[uk(t)+
Pk(t)]g2kj(t + 1) ≤ Dj , j = 1, 2, · · · , p.
(2.17)
22
Chapter 3
Power Adjustment Schemes and System Analysis
In this chapter, we develop a power control scheme for adapting the transmit power
of the secondary users [9]. The goal is to achieve the SU QoS requirement while satisfying
PU protection constraint as given in (2.14). In the rest part of the chapter, we are going to
discuss about the stability of our controller and find out our tuning point for FD and HD
modes.
3.1 PID Controller Design
First, we consider the SU QoS constraint, while ignoring the PU protection constraint.
The goal is to drive γi(t) to converge to the the SU QoS requirement Γ, for all i. The
difference between these two parameters should be considered and should be reduced as
small as possible. Another consideration is that the error signal ei(t) should be related to
the power Pi(t), which is the parameter that we need to determine for each TRi. Therefore
Pi(t) is used as the reference input. As we can see, the ratio of Γ and γi(t) can be an indicator
for the control error, and γi(t) ∝ Pi(t) if all other parameters remain the same. Thus, we
could use Γγi(t)
Pi(t) as the feedback. The error ei(t) should be the difference of feedback and
Pi(t) and we have the diagram of the PID controller as in Fig. 3.1.
The PID controller collects the SINR of each TR at every time slot and uses it as
feedback for the controller. For each time slot, let zi(t) denote the power increment from
23
Σ1)-γ(t
Γ
-
-
---
+
+
+
+
+
-+
)( tPi
-
Σ
Σ
Figure 3.1: The PID controller design.
Pi(t) to Pi(t+ 1). With feedback Γγi(t)
Pi(t), the PID controller controls the system as
ei(t− 1) =
{
Γ
γi(t− 1)− 1
}
Pi(t) (3.1)
xi(t− 1) = xi(t− 2) + ei(t− 1) (3.2)
zi(t) = KPei(t− 1) +KIxi(t− 1) +
KD |ei(t− 1)− ei(t− 2)| , (3.3)
where ei(t− 1), xi(t− 1) and |ei(t− 1)− ei(t− 2)| represent the proportional, integral and
derivative parts, respectively; Kp, KI , and KD are the corresponding coefficients. Proper co-
efficients should be designed to achieve a stable and convergent control process for adjusting
the Pi(t)’s to achieve the required minimum SINR Γ for each SU [4].
24
3.2 Power Adjustment Constraint
Next we take into account the PU protection constraint. The objective of this constraint
is to prevent the SU transmission powers from violating the interference tolerance at the DPs.
This constraint actually represents a relationship between Pi(t) and Dj , for all i and j.
We first introduce the following two parameters.
Dmin = minj=1,2,··· ,p
Dj (3.4)
ymax(t− 1) = maxj=1,2,··· ,p
yj(t− 1). (3.5)
Dmin is the minimum tolerance value among all the DPs, and ymax(t) is the maximum
measured interference among all DPs. Since Dmin is a constant and ymax(t− 1) ∝ Pi(t− 1),
the additional power constraint should also be proportional Pi(t − 1). We follow a similar
approach as in prior work [17] to introduce the following additional constraint on the power
adjustment zi(t).
ci(t) = θ(t)Pi(t− 1)− Pi(t), (3.6)
where,
θ(t) =Dmin
ymax(t− 1). (3.7)
According to (3.6) and (3.7), once the maximum interference ymax(t − 1) exceeds the
minimum tolerance Dmin, the constraint will reduce the transmit power with a proportion
of θ(t), which will drive the maximum interference back to Dmin.
Eqn. (3.6) enforces an additional constraint to the power increment zi(t) for the SUs,
so as to satisfy the PU protection constraint as given in (2.4). Because the PU protection is
a fundamental condition for spectrum sharing, the the constraint ci(t) cannot be exceeded.
25
TR� TR�+1
ConstraintConstraintConstraintConstraint
PID
ConstraintConstraintConstraintConstraint
PID
2
i
σ
� p� ()h�� +1
�=1,�≠�
2
1+iσ
� p� ()h� ,�+1 +1
�=1,�≠�+1
- -
Γ
+ +
γ �(−
1) γ�+1 (−
1)
z�() z�+1()
u�()
+
+
P�(+1)
P�()
+
+
P�+1(+1)
P�+1()
DP1
g1� g1,�+1
DP�
g� � g� ,�+1
DP�
g�� g�,�+1
�, �
u�+1()
Figure 3.2: System control block diagram.
Therefore, we have the final allowed power increment ui(t) in time slot t for TRi as
ui(t) = min{zi(t), ci(t)}, i = 1, 2, · · · , s+ 1. (3.8)
With such adjustment, the transmit power can be limited in a safe range that does not lead
to severe interference to the primary network, while trying to achieve the minimum required
SINR for the SUs. The overall diagram of the proposed power controller is illustrated in
Fig. 3.2.
26
3.3 Analysis of Stability
It is important to analyze the stability performance of the proposed power control
scheme. The stability of the PID controller (i.e., without considering FD and the PU pro-
tection constraint (3.6)) has been studied in [9]. The stability of the overall scheme depends
on the parameter settings. In the following, we examine two cases when each of the two
constraints, i.e., zi(t) and ci(t), becomes the dominant factor at the beginning stage.
Case I: zi(t) > ci(t) Initially
From (3.1), (3.2) and (3.3), we have Pi(0) = Pi(1) = Pi(2), and xi(0) = ei(0) = 0 for
the initial time slots. There is no power adjustment in the first time slot, and the first power
adjustment occurs at t = 2, as
zi(2) = (KP +KI +KD)ei(1). (3.9)
If zi(2) > ci(2), from (3.8) we have ui(2) = ci(2) and
Pi(3) = θ(2)Pi(1)
Pi(4) = θ(3)Pi(2).(3.10)
After the power adjustment in time slot 2, the detected total SU interference at DP j
in time slot 3 is
yj(3) =
s+1∑
i=1
Pi(3)g2ij =
s+1∑
i=1
θ(2)Pi(1)g2ij
=Dmin
ymax(1)
s+1∑
i=1
Pi(1)g2ij.
27
The maximum measured interference among all the DPs is
ymax(3) = maxj=1,··· ,p
yj(3) =Dmin
ymax(1)max
j=1,··· ,p
s+1∑
i=1
Pi(1)g2ij
=Dmin
ymax(1)ymax(1) = Dmin. (3.11)
Therefore the maximum measured interference will remain at Dmin and the constraint ci(t)
will remain at 0 starting from time slot 3. According to (3.8), ui(t) will also remain at 0 after
time slot 3. All the transmit powers converge to the steady value and the primary goal of PU
protection is satisfied. However, there is no guarantee that the target SU QoS requirement
can be achieved by the converged TR powers. If γi(3) < Γ, the SU QoS requirement cannot
be satisfied since the transmit powers cannot be adjusted anymore.
If zi(t) remains non-negative, ui(t) will always be 0 since ci(t) = 0 for all t ≥ 3. All
the TR powers will remain the same and the maximum measured interference remains at
Dmin. Otherwise, if zi(t) < 0 due to some disturbance, the TRi power will be reduced with
zi(t), until the target SINR Γ is reached. However, if the above two situations both happens
during the control process, we can predict that there will be oscillation and the system will
enter a bounded oscillation state. Therefore, the system can be called bounded-in-bounded-
out (BIBO) stable. In summary, for all the three cases discussed above, the system will be
stabilized by the proposed power control scheme.
Case II: zi(t) < ci(t) Initially
On the other hand, if the control function is initially dominated by the PID controller
adjustment zi(t), the pattern changes. If the transmit powers to achieve the desired SINR
cause a smaller measured interference than the Dj’s, that is, if the primary network has high
interference tolerance Dj ’s, the additional constrain enforced by ci(t) can be ignored, and
the power control will become a stable PID control process. The stability of such a system
has been demonstrated in [9].
28
However, if the desired SINR Γ cannot be achieved due to a small Dmin, the PU pro-
tection constraint will take over the control during the process and will drive the TR powers
to the maximum allowed value. The other situation is that due to the impact of some
disturbance, the control process may enter the same BIBO state as discussed in Section 3.3.
3.4 HD-FD Tuning Point
Recall that the SIS factor χ depends on the particular SIS design and is a small value in
[0, 1]. Clearly, χ, along with other network dynamics such as the channel gains, the number
and locations of SUs and DPs, and the prescribed control goals (i.e., Γ and Dj’s), all have
big impact on the system performance. So in a practical underlay CR network, it is not
true that FD transmissions will always achieve a better performance; when χ is large, the
residual SI will be so large that HD transmissions will be a better choice. Therefore, a hybrid
scheme that can switch between FD and HD modes depending on the system parameters
and states would be highly desirable. In the following, we investigate the condition under
which a switching between HD and FD modes should be made.
We use Shannon’s capacity to approximate the throughput of an SU, i.e., C = B log2(1+
SINR). Since bandwidth B is a constant for all SUs, we use the spectrum efficiency log2(1+
SINR) for comparing the efficiency of the two operation modes in the following. Let γFDi and
γHDi denote the SINRs of TRi in the FD mode and HD mode, respectively. We can derive
the average throughput for the SU pair in the HD mode, denoted as RHDi , as follows.
RHDi =
12
[
log2(1 + γHDi ) + log2(1 + γHD
i+1)]
, i is odd
12
[
log2(1 + γHDi ) + log2(1 + γHD
i−1)]
, i is even,(3.12)
where,
γHDi =
Pi+1(t)h2i+1,i
∑s+1
j=1,j 6=iPj(t)h2
ji+σ2i (t)
, i is odd
Pi−1(t)h2i−1,i∑s+1
j=1,j 6=iPj(t)h2
ji+σ2i (t)
, i is even.(3.13)
29
In the FD mode, the throughput for the SU pair is
RFDi =
log2(1 + γFDi ) + log2(1 + γFD
i+1), i is odd
log2(1 + γFDi ) + log2(1 + γFD
i−1), i is even,(3.14)
where γFDi is given in (2.6).
In each time slot t, we estimate the expected throughput for each SU pair in both the
FD and HD modes and decide which mode to adopt for the time slot. The cross-over point
for the two modes is derived by solving the following equation.
RHDi = RFD
i . (3.15)
It can be seen that (3.15) can be rewritten as
√
(1 + γHDi )(1 + γHD
i+1) = (1 + γFDi )(1 + γFD
i+1),
i is odd√
(1 + γHDi )(1 + γHD
i−1) = (1 + γFDi )(1 + γFD
i−1),
i is even.
(3.16)
Define a ratio T as follows.
Ti =
√(1+γHD
i )(1+γHDi+1
)
(1+γFDi )(1+γFD
i+1), i is odd
√(1+γHD
i )(1+γHDi−1
)
(1+γFDi )(1+γFD
i−1), i is even.
(3.17)
Thus, we have the following proposition for determining the operation mode for TRi in the
hybrid scheme.
Proposition 1. A TRi should operate in the HD mode if Ti ≥ 1, and it should operate in
the FD model if Ti < 1.
30
Chapter 4
Simulation and Analysis
4.1 Simulation Configuration
To evaluate the performance of the proposed power control scheme for FD underlay
CR networks, we conduct extensive simulations using a MATLAB implementation. We use
one network in a 100×100 area and another network in a 1000×1000 area. The outdoor
channel model h = 40 log10(d) + 10 dB is used in all the simulations, where d is the distance
between the transmitter and receiver. In each simulation, the noises powers σ2 are i.i.d.
random variables evenly distributed in a fixed range, while the range may change in different
simulations. As discussed, the performance of FD systems are greatly affected by χ. We
choose χ = 0.00005 in most of the simulations, unless otherwise specified.
There are eight TRs and four DPs in the network. The location of the TRs and DPs
are shown in Fig. 4.1 and Fig. 4.13. We assume each TR can communicate with all the DPs
through a control channel to obtain information about detected interference level at the DPs
(i.e., ymax(t− 1)). The control goals Γ and Dj ’s are prescribed and is known to all the TRs.
Such information is used as input to the control scheme executed at each TR to adjust its
transmit power.
4.2 Control Performance Analysis
In this section, we evaluate the performance of the proposed controller in the FD and HD
modes. First, we simulate the 100×100 network under fixed Γ and fixed χ. We set Γ = 0.1
and χ = 0.00005 in the simulation. Noise σ2 is uniform distributed in [1.2×10−6, 2.4×10−8]
W. DP’s tolerance limit is set as Dj = 5× 10−10W for all j.
31
0 20 40 60 80 1000
10
20
30
40
50
60
70
80
90
100
x
y
TR locationLocation of TR5Detection point location
Figure 4.1: Distribution of the TRs and DPs in the 100×100 network.
Take TR5 as an example. The evolutions of its SINR γ5(t) and transmit power P5(t) are
plotted in Fig. 4.2 and Fig. 4.3, respectively. It can be seen that both the SINR and power
curves quickly converge to the neighborhood of the stable values, and then fluctuate around
the stable values. In Fig. 4.2, the HD SINR is slightly higher than that of HD. However this
does not mean that the throughput of an FD link is lower than that of an HD link, since
the FD link throughput is the sum of that of the two end TRs. In Fig. 4.3, it can be seen
that a higher transmit power is used in the FD mode to overcome the residual SI in order
to achieve the target SINR value Γ.
Due to the parameters used, in the HD mode the power adjustment of TR5 is initially
dominated by the PU protection constraint (3.6), and then controlled by the SU QoS con-
straint (3.3). In the FD case the control of the power adjustment is jointly done by both
constraints. This can be witnessed by comparing u5(t) = min{z5(t), c5(t)}, z5(t), and c5(t)
as plotted in Fig. 4.4, Fig. 4.5, and Fig. 4.6, respectively. In a few time slots the control
32
0 50 100 150 2000
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Time
SIN
R o
f TR
5
ΓFD modeHD mode
Figure 4.2: Evolution of the SINR at TR5 when Γ = 0.1 and χ = 0.00005.
0 50 100 150 2000
0.2
0.4
0.6
0.8
1x 10
−3
Time
Pow
er o
f TR
5 (w
)
FD modeHD mode
Figure 4.3: Evolution of the transmit power at TR5 when Γ = 0.1 and χ = 0.00005.
33
0 20 40 60 80 100 120 140 160 180 200−10
−8
−6
−4
−2
0
2
4x 10
−4
Time
Pow
er a
djus
tmen
t of T
R5
(W)
FD modeHD mode
Figure 4.4: Power adjustment u5(t) = min{z5(t), c5(t)} of TR5 when Γ = 0.1 and χ =0.00005.
process u5(t) reaches the stable value 0 and achieves the optimal SINR with the given D for
TR5.
In Fig. 4.7, we demonstrate the PU protection performance by plotting the PU inter-
ference tolerant D and the maximum measure interference at the DPs for the FD and HD
modes. It can be seen that with the proposed power control scheme, the maximum DP
detected interference ymax defined in (3.7) quickly drops below D after a few time slots.
Therefore the PU protection goal is well achieved by the proposed power control scheme. In
the meantime, the controlled power remains around 0.4 W for the FD mode and 0.2 W for
the HD mode (see Fig. 4.3), which are sufficient to satisfy the required SINR Γ = 0.1 for TR5,
as shown in Fig. 4.2. Since the controlled power of TR5 has achieved both PU protection
and SU QoS goals, the power adjustment u5(t) will stay within a narrow range around 0.
Next we try a large value of Γ = 0.3 in the simulation to evaluate the case of an overly
high QoS requirement of the SUs that cannot be supported in the underlay CR network.
That is, there is no feasible solution to satisfy both PU protection and SU QoS constraints
34
0 50 100 150 200−4
−3
−2
−1
0
1
2
3
4
5x 10
−4
Time
PID
incr
emen
t of T
R5
(W)
FD modeHD mode
Figure 4.5: PID controller adjustments z5(t) when Γ = 0.1 and χ = 0.00005.
0 20 40 60 80 100 120 140 160 180 200−10
−8
−6
−4
−2
0
2
4x 10
−4
Time
PU
pro
tect
ion
incr
emen
t of T
R5
(W)
FD modeHD mode
Figure 4.6: PU protection constraint adjustments c5(t) when Γ = 0.1 and χ = 0.00005.
35
0 50 100 150 2000
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6x 10
−8
Time
Max
imum
DP
det
ecte
d in
terf
eren
ce (
W)
DP tolerance DFD modeHD mode
Figure 4.7: Maximum measured interference among all the DPs when Γ = 0.1 and χ =0.00005.
in this case. In this case, the power control scheme will try to achieve PU protection as a
primary goal, as the prerequisite condition for spectrum sharing, and then try to maximize
the SINR of the SUs as a secondary goal. The maximum DP detected interferences to PUs
are plotted in Fig. 4.8 for the FD and HD modes. It can be seen that the proposed power
control scheme can effectively guarantee that the interference to PUs is below the tolerance
D. The achieved SINR for TR5 is also plotted for the FD and HD modes in Fig. 4.9. It can
be seen that the SINR of TR5 is stabilized around 0.17 for the HD mode and 0.1 for the FD
mode, although the desired SINR for TR5 is 0.3. Since the maximum allowed interference
has been reached (i.e., D = 5 × 10−10 W as in Fig. 4.8), the power of TR5 cannot be
further increased to reach the target SINR level Γ = 0.3. Note that although the HD mode
achieves a higher SINR than the FD mode as in Fig. 4.9, this does not necessarily mean
the HD throughput is higher than that of the FD model. We will examine the throughput
performance in Section 4.3.
36
0 20 40 60 80 100 120 140 160 180 2000
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6x 10
−8
Time
Max
imum
DP
det
ecte
d in
terf
eren
ce (
W)
DP tolerance DFD modeHD mode
Figure 4.8: Maximum measured interference among all the DPs when Γ = 0.3 and χ =0.00005.
0 20 40 60 80 100 120 140 160 180 2000
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Time
SIN
R o
f TR
5
ΓFD modeHD mode
Figure 4.9: Evolution of the SINR at TR5 when Γ = 0.3 and χ = 0.00005.
37
Table 4.1: Setting of Γ.
Time slot (t) 1-50 51-100 101-150 151-200
Fig. 4.10 Γ 0.1 0.05 0.05 0.1
Fig. 4.11 Γ 0.05 0.1 0.15 0.2
Finally, we demonstrate the performance of the proposed power controller under vary-
ing SU QoS requirements. In particular, we vary the SU SINR requirement Γ as given in
Table 4.2. In Fig. 4.10, the required SINR is with the range such that the proposed controller
can achieve both PU protection and SU QoS goals. It can be seen that the SINR can be
stabilized around the target SINR for the full range. In Fig. 4.11, the SINR is continuously
increased, from the feasible range to the infeasible range both goals cannot be met simul-
taneously. In Fig. 4.11, the HD SINR curve is controlled well and the SINR of TR5 can
quickly follow Γ from 0.05 to 0.2. On the other hand, the FD SINR curve cannot follow the
increased Γ byond 0.1, because of a larger power is required to combat the residual SI in
order to achieve the same SINR, which, however, is not allowed since the primary constraint
of PU protection will be violated. In Fig. 4.12, the maximum measured interference among
the DPs is also plotted. It can be seen that the primary goal of PU protection is always
achieved by the proposed scheme.
4.3 Throughput Performance
In this section, we evaluate the achievable throughput by the proposed power control
scheme. We focus on the proposed hybrid scheme in the simulations, with which the operat-
ing mode for each TR is determined as given in Section 3.4. The large 1000×1000 network
with eight TRs and four DPs is used in the following simulations, as shown in Fig. 4.13.
In the simulation, we increase χ from 0.000005 to 0.005 with step size of 0.000005. With
each χ value, we simulate the system for 200 time slots each with a random noise level, which
has been shown to be sufficient long for convergence in our previous simulations. We plot
38
0 50 100 150 2000
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Time
SIN
R o
f TR
5
ΓFD modeHD mode
Figure 4.10: Evolution of the SINR at TR5 under varying Γ.
0 20 40 60 80 100 120 140 160 180 2000
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Time
SN
R o
f TR
5
ΓFD modeHD mode
Figure 4.11: Evolution of the SINR at TR5 under varying Γ.
the average throughput of the 200 time slots for each χ value in the figure for the FD, HD
and hybrid schemes.
39
0 50 100 150 2000
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6x 10
−8
Time
Max
imum
DP
det
ecte
d in
terf
eren
ce (
W)
DP tolerance DFD modeHD mode
Figure 4.12: Maximum measured interference among all the DPs under varying Γ.
0 200 400 600 800 10000
100
200
300
400
500
600
700
800
900
1000
x
y
TR locationDetection point location
Figure 4.13: Distribution of the TRs and DPs in the 1000×1000 network.
40
0 1 2 3 4 5
x 10−3
0
0.5
1
1.5
2
2.5
χ
Ave
rage
Thr
ough
put (
bits
/sec
/Hz)
FD modeHD modeHybrid mode
Figure 4.14: Average throughput of the FD, HD, and hybrid modes for different values of χ.
It can be seen in Fig. 4.14 that the hybrid scheme achieves the highest throughput for the
entire range of χ. In particular, when χ ≤ 3.43× 10−4, SIS is very effective and most of the
TRs operate in the FD mode. The hybrid scheme achieves the same throughput as FD, which
is higher than that of HD. As χ is increased, the advantage of FD transmissions diminishes
and HD begins to achieve higher throughput than FD. When 3.43× 10−4 ≤ χ ≤ 1.3× 10−3,
some TRs operate in the FD mode and some others in the HD mode. When χ ≥ 1.3× 10−3,
all the TRs operate in the HD mode since the residual SI is so strong, there is no benefit
for using FD transmissions. The proposed hybrid scheme compares the gains of FD and
HD, and always chooses the better operating mode to achieve the highest throughput for
the entire range of χ.
Finally, we investigate the impact of noise level. In the simulation, we set χ to 0.0005
and increase the noise power σ2 from 10−6W to 10−3W . The throughput results for the
three schemes are presented in Fig. 4.15. As expected, the hybrid scheme achieves the
highest throughput among the three, and the throughput decreases when noise is increased
41
0 0.2 0.4 0.6 0.8 1
x 10−3
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Average Noise (W)
Ave
rage
Thr
ough
put (
bits
/sec
/Hz)
FD modeHD modeHybrid mode
Figure 4.15: Average throughput of the FD, HD, and hybrid modes for different values ofσ2.
for all the three schemes. However, the influence of noise on throughput is different for the
three schemes.
As shown in (2.6), the FD mode has one extra interference source, i.e., the residual
SI, making it less sensitive to the varying noise power. This is why the throughput of HD
decreases faster than FD. The hybrid scheme can use FD instead of HD even though the χ
value is larger in this simulation. The hybrid scheme always achieves the highest throughput
in all the scenarios simulated.
42
Chapter 5
Conclusions and Future Work
5.1 Conclusion
In this paper, we investigate the design of effective power controllers for single channel
underlay CR networks, where FD transmissions are exploited to improve the network capac-
ity. Taking the SIS factor into consideration, we investigate the design of a PID controller
and a hybrid FD-HD scheme to achieve the dual goals of PU protection and SU QoS provi-
sioning. The stability performance of the proposed scheme is analyzed and evaluated with
simulations.
In the previous chapters, we developed a hybrid mode power controlled transmission
system over underlay CR network, which enables SUs achieving optimum throughput while
remaining undetected from CR PUs. Our research includes PID controller design, network
optimization, performance analysis and simulation validation.
In the begining of Chapter 2 we explicitly stated our system, which is very practical in
reality. In the following, we also discussed the problem, which is a underlay CR network
based power control and throughput optimization quest.
Then, we designed our general PID controller in Chapter 3 to control transmission power
in order to gain optimal throughput. However the PID controller could not guarantee the
control systems stealth ability, so we set a constraint to every power adjustment to enable
such function.
Then, during researching on our system, we analyzed the stability of the system. Mean-
while we found simply using FD or HD mode would not provide us with optimal performance,
or in another word, throughput. So we also tackled with this more challenging problem and
43
found the tuning point of the two modes. Finally we combined our former proposed system
with the hybrid mode to gain a better throughput than any of the former two modes.
In the end, we had lots of detailed and accurate simulations under different environments
in Chapter 4. The results which has been shown by the results suggested both stability and
undoubtable superior performance of our system.
5.2 Future Work
Although our proposed system has already shown considerable advantages comparing to
original CR network, there are still many topics to be explored which can make our system
even better. For example, maybe we could design an optimization algorithm to allocate
the transmission power for each SU to gain a even better, organized overall throughput;
We could also do more work on the constraint, which may degrade the performance of PID
controller. The optimization and the reestablishment of constraint could even be studied
together to create a dynamic condition which assures better efficiency of the system.
Also, certain validation may still be needed. Since we have only simulated the system
using software, the real performance may not be guaranteed. Thus, further verification like
testbed based hardware simulation should be performed in order to check the availability in
reality. Throughout the booming development of FD hardware, testbed validation which is
much more convincing should definitely be our next target.
Last but not the least, our proposed system shows some potential ability of combining
with other technology. For instance, relay network, which need fairness as well as optimal
throughput. The problem of general optimization we’ve mentioned before could be studied
together with relay network and may generate some interesting tradeoff during researching.
If such kind of tasks we have listed before can be accomplished, we are sure that the
system will present an even better availability and efficiency.
44
Bibliography
[1] Joseph Mitola III. http://web.it.kth.se/~maguire/jmitola/.
[2] SDR product. http://www.flex-radio.com/Data/Image/sdr1k-enc_large.jpg.
[3] Wessam Afifi and Marwan Krunz. Exploiting self-interference suppression for improvedspectrum awareness/efficiency in cognitive radio systems. In Proc. IEEE INFOCOM’13,pages 1258–1266, Turin, Italy, 2013.
[4] Kiam Heong Ang, Gregory Chong, and Yun Li. PID control system analysis, design,and technology. IEEE Trans. Control Syst. Technol., 13(4):559–576, July 2005.
[5] Ehsan Aryafar, Mohammad Amir Khojastepour, Karthikeyan Sundaresan, SampathRangarajan, and Mung Chiang. MIDU: enabling MIMO full duplex. In Proc. ACMMobiCom’12, pages 257–268, Istanbul, Turkey, Aug. 2012.
[6] Jung Il Choi, Mayank Jain, Kannan Srinivasan, Phil Levis, and Sachin Katti. Achievingsingle channel, full duplex wireless communication. In Proc. ACM MobiCom’10, pages1–12, Chichago, IL, Sept. 2010.
[7] Melissa Duarte and Ashutosh Sabharwal. Full-duplex wireless communications usingoff-the-shelf radios: Feasibility and first results. In Proc. ASILOMAR’10, pages 1558–1562, Pacific Grove, CA, Nov. 2010.
[8] S. Goyal, P. Liu, S. Hua, and S.S. Panwar. Analyzing a full-duplex cellular system. InProc. CISS’13, pages 1–6, Baltimore, MD, Mar. 2013.
[9] Fredrik Gunnarsson, Fredrik Gustafsson, and Jonas Blom. Pole placement design ofpower control algorithms. In IEEE VTC-Spring’99, pages 2149–2153, Houston, TX,May 1999.
[10] Seishi Hanaoka, Masashi Yano, and Shinji Nishimura. Platform for inter-radio systemswitching with cognitive radio. 2009.
[11] Simon Haykin. Cognitive radio: brain-empowered wireless communications. IEEE J.Sel. Areas Commun., 23(2):201–220, Feb. 2005.
[12] Y. Huang, S. Mao, and S. F. Midkiff. A control-theoretic approach to rate control forstreaming videos. IEEE Trans. Multimedia, 11(6):1072–1081, Oct. 2009.
[13] IEEE. IEEE 802.22-2011(TM) Standard for Cognitive Wireless Regional Area Networks(RAN) for Operation in TV Bands, July 2011.
45
[14] Mayank Jain, Jung Il Choi, Taemin Kim, Dinesh Bharadia, Siddharth Seth, KannanSrinivasan, Philip Levis, Sachin Katti, and Prasun Sinha. Practical, real-time, fullduplex wireless. In Proc. ACM MobiCom’11, pages 301–312, Las Vegas, NV, Sept.2011.
[15] Sarah Koskie and Zoran Gajic. Optimal SIR-based power control strategies for wirelessCDMA networks. J. Inform. And Syst. Sciences, 1(1):1–18, Jan. 2007.
[16] Ying-Chang Liang, Yonghong Zeng, Edward CY Peh, and Anh Tuan Hoang. Sensing-throughput tradeoff for cognitive radio networks. IEEE Trans. Wireless Commun.,7(4):1326–1337, Apr. 2008.
[17] Genki Matsui, Takuji Tachibana, Yukinori Nakamura, and Kenji Sugimoto. Distributedpower adjustment based on control theory for cognitive radio networks. Elsevier Com-puter Netw., 57(17):3344–3356, Dec. 2013.
[18] U.S. Department of Commerce, National Telecommunications, Information Adminis-tration, and Office of Spectrum Management. United states frequency allocations–theradio spectrum. 10 2003.
[19] Bozidar Radunovic, Dinan Gunawardena, Peter Key, Alexandre Proutiere, Nikhil Singh,Vlad Balan, and Gerald Dejean. Rethinking indoor wireless mesh design: Low power,low frequency, full-duplex. In Wireless Mesh Networks (WIMESH 2010), 2010 FifthIEEE Workshop on, pages 1–6. IEEE, 2010.
[20] Achaleshwar Sahai, Gaurav Patel, and Ashutosh Sabharwal. Pushing the limits of full-duplex: Design and real-time implementation. arXiv preprint, arXiv:1107.0607, 2011.
[21] X. Shen, X. Cheng, L. Yang, M. Ma, and B. Jiao. On the design of the schedulingalgorithm for the full duplexing wireless cellular network. In Proc. IEEE Globecom’13,pages 1–6, Atlanta, GA, Dec. 2013.
[22] Alexander M Wyglinski, Maziar Nekovee, and Thomas Hou. Cognitive radio communi-cations and networks: principles and practice. Academic Press, 2009.
[23] Qing Zhao and Brian M Sadler. A survey of dynamic spectrum access. IEEE SignalProcess. Mag., 24(3):79–89, May 2007.
[24] Y. Zhao, S. Mao, J. Neel, and J. H. Reed. Performance evaluation of cognitive radios:metrics, utility functions, and methodologies. Proc. IEEE, 97(4):642–659, Apr. 2009.
46